Autocalibrated multi-shot magnetic resonance image reconstruction with joint optimization of shot-dependent phase and parallel image reconstruction

Abstract

Images are reconstructed from k-space acquired with a magnetic resonance imaging (“MRI”) system using a multi-shot pulse sequence. In each iteration, a phase-aware image reconstruction, a data-consistency update across all shots or subsets of data, and a relative phase estimation across the reconstructed images for each shot are performed. In this way, the reconstruction framework recasts the problem as an iterative relative phase estimation problem, which allows for the use relative phase estimation techniques. Through an iterative search, an artifact-free combined image and the smooth relative phase between each shot in the multi-shot k-space data can be jointly estimated.

The invention claimed is: 
     
         1 . A method for reconstructing an image from data acquired with a magnetic resonance imaging (MRI) system, the method comprising:
 (a) accessing multi-shot k-space data with a computer system, wherein the multi-shot k-space data have been acquired with an MRI system using a pulse sequence comprising multiple shots;   (b) accessing shot-dependent phase data with the computer system, wherein the shot-dependent phase data comprise estimates of relative phase differences between the multiple shots;   (c) reconstructing, with the computer system, an image of the subject from the multi-shot k-space data using an image reconstruction algorithm that incorporates and iteratively updates the shot-dependent phase data based in part on a self-consistency of the reconstructed image with the multi-shot k-space data.   
     
     
         2 . The method of  claim 1 , wherein accessing the shot-dependent phase data include estimating initial shot-dependent phase data with the computer system by reconstructing initial images from the multi-shot k-space data and estimating the initial shot-dependent phase data from the initial images. 
     
     
         3 . The method of  claim 1 , wherein the shot-dependent phase data are updated in each iteration of the image reconstruction algorithm using an ESPIRIT algorithm to estimate an updated relative phase differences between shots of the multi-shot k-space data. 
     
     
         4 . The method of  claim 1 , wherein the image reconstruction algorithm comprises, in each iteration:
 reconstructing coil image data comprising a coil image for each coil in a multicoil receive array for each of the multiple shots in the multi-shot k-space data;   forming a coil combined image for each shot by combining coil images in the coil image data for each shot;   evaluating a data consistency condition between the combined coil image for each shot and the multi-shot k-space data.   
     
     
         5 . The method of  claim 1 , wherein the multi-shot k-space data are undersampled multi-shot k-space data. 
     
     
         6 . The method of  claim 5 , wherein the undersampled multi-shot k-space data are undersampled in at least one in-plane k-space dimension. 
     
     
         7 . The method of  claim 1 , wherein a machine learning (ML) prior is incorporated into the image reconstruction algorithm. 
     
     
         8 . The method of  claim 1 , wherein accessing the multi-shot k-space data with the computer system includes acquiring the multi-shot k-space data with the MRI system and accessing the acquired multi-shot k-space data. 
     
     
         9 . The method of  claim 1 , wherein the multi-shot k-space are acquired using a multi-shot echo planar imaging (EPI) pulse sequence. 
     
     
         10 . The method of  claim 4 , wherein the image reconstruction algorithm is stopped when evaluating the data consistency condition satisfied a stopping criterion. 
     
     
         11 . The method of  claim 6 , wherein the undersampled multi-shot k-space data are undersampled in a phase-encoding dimension. 
     
     
         12 . The method of  claim 7 , wherein the ML prior comprises a network that has been trained on training data to minimize artifacts attributable to model errors.